Volume 9, Issue 10, October – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT580

Overview of Cyber Attacks Classification and

Detection in IoT using CNN-Deep
Reinforcement Learning
Katikam Mahesh1*
Scholar at Andhra University College of Engineering (AUCE(A), Visakhapatnam-
530003, Andhra Pradesh, India, in the Department of Computer Science and Systems Engineering
https://orcid.org/0009-0000-0707-1117

Dr. Kunjam Nageswara Rao2

Professor at Andhra University College of Engineering (AUCE(A), Visakhapatnam –
530003, Andhra Pradesh, India, in the Department of Computer Science and Systems Engineering.
https://orcid.org/ 0009-0005-2779-0238

Coresopinding Author: Katikam Mahesh1*

Abstract:- Millions of digital devices total the Internet of II. CLASSIFICATION OF CYER
Things (IoT), and this allows very easy interaction from ATTACKS IN IOT
users connecting the devices. IoT is one of the tech
sectors that is expanding most rapidly, but it can also be A. Physical Attacks
very vulnerable to hazards. Infections and abnormal IoT device hardware be the target of physical attacks.
placement on the Internet of Things (IoT) framework is Attackers may trigger malfunctions or criminal data access
an increasing threat in the field of technology. In view of via tampering devices, messing with sensors, or having
the growing IoT foundation usage across all industries, unauthorized physical access to the devices.
attacks and dangers on these systems have also grown
proportional. Leveraging typical machine learning  Zero-Day Attacks
methods, cyber-attack detection plays a critical role in This attack subtype relies on an undetected flaw in IoT
avoiding damage from cyberattacks on IoT devices. IoT devices or software. These vulnerabilities can be highly
Cyber Attacks are Not Detected by ANN Artificial effective and tough to guard against because attackers make
Neural Networks Using Deep Learning Techniques use of them before developers have a chance to issue
CNN-DRL (Convolutional Neural Networks-Deep patches.
Reinforcement Learning) Hybrid Approach: Detects
Attacks, including Distributed Denial of Service (DDoS),  Eavesdropping Attacks
Zero-day, and Eavesdropping Attacks. Attackers maintain a watch on the way IoT devices
and communication channels interact with others, with the
Keywords:- Cyber Attacks, Internet of Things (IoT), goal of stealing user credentials or sensitive data. To counter
Convolutional Neural Networks, Deep Reinforcement this threat, secure communication protocols and proper
Learning. encryption are required.

I. INTRODUCTION  Data Injection Attacks

Malicious codes and commands are be employed by
Deep learning employs algorithms that, by examining attackers to attack poorly protected systems.
data with a logical structure, try to draw conclusions that
people would. Deep learning uses neural networks, multi-  Replay Attacks
layered structures of algorithms that can detect patterns and Through the application of malicious instructions, an
classify data in the same way as people can, to do this.Our approved info packet gets modified to carry off this attack.
brains attempt to draw similarities among newly gained Electronic equipment receive malicious packets passed on
information and what we already know. Neural networks are through them under the disguise of totally legitimate
utilized in deep learning to carry out the same idea. Big information packets.
amounts of unstructured, data without labels are all that
deep learning systems need in order to learn on their own
and produce elegant, accurate forecast models.

IJISRT24OCT580 www.ijisrt.com 569

Volume 9, Issue 10, October – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT580

B. Denial of Service (DoS) and Distributed Denial of III. RELATED WORKS

Service (DDoS) Attacks.
Attackers flood IoT systems with a great deal of The art of incorporating machine learning has been the
traffic, leaving them unresponsive or unusable, disrupting focus of numerous studies in the past and a number of
vital services and inflicting financial and reputational professional works on the use of machine intelligence and
damages. This is done by taking advantage of flaws in the data-mining techniques for intrusion detection have been
security protocols of IoT devices. published [1]. But almost all of these earlier studies have
only employed machine learning methods to identify
C. Botnet Attacks intruders in conventional networks. As a result, in this work,
Attackers take over weakly secured IoT devices to we are expanding this field of study by particularly using
build botnets. These botnets can be used for various machine learning to recognize attacks in the conditions of
malicious activities, such as sending out spam emails, Numerous studies has already concentrated on the art of
launching DDoS attacks, or spreading malware. incorporating machine learning 2], and a number of
professional works on the utilization of data-mining and
D. Man-in-the-Middle (MitM) Attacks machine intelligence methods in security detection have
Attackers can eavesdrop on sensitive data or insert mal been published [3]. However, nearly all of these prior
icious things via MitM crimes.This type of attack targets we studies have solely utilized machine learning techniques to
ak authentication, weak encryption, and unsafe network con detect intrusions via conventional networks. As a result, we
nections. extend this area of research in this work by specifically
deploying machine learning to identify attacks under certain
E. Malware Attacks Attacks conditions of
Malicious software compromising their operation and
spreading to other devices in the network. Since We looked at a number of books and compiled them in
ransomware can turn down devices and demand payment to Table I to analyze recent research on the subject of attack
release them, attacks on the web of Things, or IoT, devices detection using machine learning in IoT networks. The
are growing progressively troubling. datasets, detection strategies, and learning approaches have
been provided for each research. We explored the usage of
F. Credential Attacks various machine learning approaches and datasets when
Attackers use default or weak passwords to access IoT choosing these studies. The studies provide evidence that
devices with authority. Once inside, they change the settings machine learning techniques can achieve success for attack
of the device, take advantage of sensitive data, or use the detection. From the works discussing the issue of using
device as a stepping stone to get onto a broader network. machine learning for IoT security, the detection
methodologies can be categorized as unsupervised methods
G. Firmware Attacks [6], [7], [8], [14] and supervised methods [9], [10], [11],
By rewriting the firmware, attackers take over the [11, [12]
workings of the device and are potentially allowed to
control it remotely or provide unauthorized access.

H. Side-Channel Attacks
Attackers using side channel take employ of data that
is exposed if the device is working regularly, such as
patterns of power use or electromagnetic emissions. IoT
attacks like these have the capacity for exposing encryption
keys and other personal information.

I. Encryption Attacks
If the user's IoT device is not encrypted, then attackers
can install and modify their algorithms and take over your
device. In regard to the significance of encryption in IoT
devices, this is a crucial detail to remember.

J. Brute Force Password Attack

Software that can make a large number of password
combinations is used in this attack, and the attacker
distributes these to a targeted user base. The attack targets
accounts that have a weak password securing them. As a
result, the attacker is free to steal confidential data, spread
malware, and carry any other actions.

IJISRT24OCT580 www.ijisrt.com 570

Volume 9, Issue 10, October – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT580

Table 1: Summery of Related Works

IV. PROPOSED METHDOLOGY interesting elements from IoT network traffic data by
utilizing CNN's features. In order to identify the most useful
CNN, our suggested method for feature selection and characteristics to correctly classify network traffic as either
threat detection on Internet of Issues networks. CNN utilizes attack or regular, CNN-CNN consists of two separate CNN
the features of a dual convolutional neural network models that cooperate in some way. One model is used for
architecture. Our approach is to automatically choose feature selection, and the other is used for attack detection.

Fig 1: CNN For Attack Detection

Convolutional neural networks are distinguished from DRE deep reinforcement learning Within the field of
other neural networks by their superior performance with machine learning, a new research area is called While neural
image, speech, or audio signal inputs. They have three main networks have made recent advances in fields such as time
types of layers, which are: series, computer vision, and machine translationNeural
network deep learning. Many of DRL achievements can be
 Convolutional layer attributed to extending earlier RL work to high-dimensional
 Pooling layer situations. This is brought about by neural networks' strong
 Fully-connected (FC) layer function approximation capabilities and their ability to learn

IJISRT24OCT580 www.ijisrt.com 571

Volume 9, Issue 10, October – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT580

low-dimensional feature representations. Unlike tabular and learning (RL) agents, enabling them to learn directly from
typical nonparametric approaches, DRL may successfully high-dimensional, raw visual inputs. Training deep neural
conquer the curse of dimensionality through representation networks to approximate the optimal policy π* and/or the
learning [7]. Convolutional neural networks (CNNs), for optimal value functions V*, Q*, and A* is the broad
example, can be employed as parts of reinforcement foundation for deep learning via reinforcement (DRL).

Fig 2: DRE for Attack Classification

 Data Preprocessing To get ready for machine learning, data is collected from
The process of changing raw data into a format that is many sources and scrubbed. It might not be in a suitable
suitable to machine learning is known as data preparation. form, have sounds and missing data, or both.

Fig 3: Data Preprocessing Stages

IJISRT24OCT580 www.ijisrt.com 572

Volume 9, Issue 10, October – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT580

V. RESULTS

df= pd.read_csv(r'C:\Users\qures\Downloads\sample.csv')
print(df)

Fig 4: Data Set Importing to Classify Attacks

VI. CONCLUSION ACKNOWLEDGEMENT

The Internet of Things (IoT) is a system made up of I would like to express my profound gratitude to Dr.
millions of digital devices and people can easily interact Kunjam Nageswara Rao Professor Department of Computer
with the devices by connecting them. One of the digital Science and System Engineering in Andhra University
sectors that is expanding the fastest is IoT, but it may also Andhra Pradesh, Visakhapatnam India, for giving Guidance
be very hazardous. A increasing hazard in the sphere of and Support to Review and Given suggestion.
technology is the Internet of Things (IoT) buildings that can
be damaged by infections and faulty duty. Growing  Author Contribution
adoption of internet of things in various industries has led to All authors contributed to the study conception and
a proportional rise in attacks and dangers on these systems. design. Material preparation, data collection and analysis
Cyber-attack detection, which makes utilization of basic were performed by Katikam Mahesh, Dr Kunjam
machine learning methods, is critical for avoidance of Nageswara Rao, and all authors commented on previous
damage from cyberattacks on Iot of Things devices. I.T. versions of the manuscript.
Artificial Neural Networks (also called ANN Do Not Detect
Cyberattacks Using Deep Learning The techniques  Conflicts of Interest
Convolutional neural networks-deep reinforcement learning The authors declare no conflicts of interest
(CNN-DRL) hybrid approach: Detect attacks, such as
eavesdropping, zero-day, and distributed denial of service
(DDoS) attacks.In future Classifly more Attcks Using
Advanced Deeop Learnig Technques.

IJISRT24OCT580 www.ijisrt.com 573

Volume 9, Issue 10, October – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24OCT580

REFERENCES [14]. E. Hodo, X. Bellekens, A. Hamilton, P.-L. Dubouilh,

E. Iorkyase, C. Tachtatzis, and R. Atkinson, “Threat
[1]. G. Eason, B. Noble, and I.N. Sneddon, “On certain analysis of iot networks using artificial neural
integrals of Lipschitz-Hankel type involving network intrusion detection system,” International
products of Bessel functions,” Phil. Trans. Roy. Soc. Symposium on Networks, Computers and
London, vol. A247, pp. 529-551, April 1955. Communications (ISNCC), pp. 1–6, 2016.
(references)
[2]. J. Clerk Maxwell, A Treatise on Electricity and
Magnetism, 3rd ed., vol. 2. Oxford: Clarendon, 1892,
pp.68-73.
[3]. I.S. Jacobs and C.P. Bean, “Fine particles, thin films
and exchange anisotropy,” in Magnetism, vol. III,
G.T. Rado and H. Suhl, Eds. New York: Academic,
1963, pp. 271-350.
[4]. K. Elissa, “Title of paper if known,” unpublished.
[5]. R. Nicole, “Title of paper with only first word
capitalized,” J. Name Stand. Abbrev., in press.
[6]. Y. Mirsky, T. Doitshman, Y. Elovici, and A. Shabtai,
“Kitsune: an ensemble of autoencoders for online
network intrusion detection,” arXiv preprint
arXiv:1802.09089, 2018.
[7]. X. Yuan, C. Li, and X. Li, “Deepdefense: identifying
ddos attack via deep learning,” IEEE International
Conference on Smart Computing (SMARTCOMP),
pp. 1–8, 2017.
[8]. Y. Meidan, M. Bohadana, Y. Mathov, Y. Mirsky, A.
Shabtai, D. Breitenbacher, and Y. Elovici, “N-
baiot—network-based detection of iot botnet attacks
using deep autoencoders,” IEEE Pervasive
Computing, vol. 17, no. 3, pp. 12–22, 2018.
[9]. M. A. Ferrag and L. Maglaras, “Deepcoin: A novel
deep learning and blockchain-based energy exchange
framework for smart grids,” IEEE Transactions on
Engineering Management, 2019.
[10]. Y. Meidan, M. Bohadana, A. Shabtai, M. Ochoa, N.
O. Tippenhauer, J. D. Guarnizo, and Y. Elovici,
“Detection of unauthorized iot devices using
machine learning techniques,” arXiv preprint
arXiv:1709.04647, 2017.
[11]. N. Koroniotis, N. Moustafa, E. Sitnikova, and J.
Slay, “Towards developing network forensic
mechanism for botnet activities in the iot based on
machine learning techniques,” International
Conference on Mobile Networks and Management,
pp. 30–44, 2017.
[12]. Y. N. Soe, Y. Feng, P. I. Santosa, R. Hartanto, and
K. Sakurai, “Rule generation for signature-based
detection systems of cyber-attacks in iot
environments,” Bulletin of Networking, Computing,
Systems, and Software, vol. 8, no. 2, pp. 93–97,
2019.
[13]. V. H. Bezerra, V. G. T. da Costa, S. B. Junior, R. S.
Miani, and B. B. Zarpelao, “One-class classification
to detect botnets in iot devices,” Anais do XVIII
Simposio Brasileiro em Seguranc¸a da Informac¸ ´ ao
e ˜ de Sistemas Computacionais, pp. 43–56, 2018.

IJISRT24OCT580 www.ijisrt.com 574